Re: Missing ranked data
- From: sxyvirgo@xxxxxxxxx
- Date: 15 Sep 2006 06:04:57 -0700
Thom -
I may not have explained this well - participants will be ranking
websites so if they only see two of the three, so I guess replacing the
missing data with the average ranking will be 1.5 for every person?
Each website has an equal probability of being last, and thus skipped
for lack of time because of the counterbalancing used.
And I'd much rather use regular ANOVA just because of the simplicity
and flexibility if appropriate....so that's good.
Thom wrote:
sxyvirgo@xxxxxxxxx wrote:
The issue is that it's quite possible some persons will take too long
and will only be able to see two of the three websites. In these
cases, we'll still have the measures they completed for the individual
websites but does it make any sense at all to have them complete the
ranking items for the two websites they did get to see? In other
words, is there any way to appropriately use those incomplete rankings
or will that data need to be discarded (or not even collected) if it
can't include all three of the websites?
Yes - but the methods are quite sophisticated (e.g., multiple
imputation or maybe something like multilevel model). Discarding cases
will almost certainly introduce bias unless data are missing completely
at random (MCAR) which is very unlikely. The bias will be small if the
discarded cases are relatively few in number.
I'd planned on using Friedman's ANOVA for ranks. So if these cases
Why? If the raw data are alreaddy ranks then you don't need to use a
rank transformation test - standard ANOVA will do fine (and may be much
more flexible).
appear to be randomly distributed across the conditions, would it make
sense to merely give a ranking of "3" for that website for all those
items?
No. That would add bias and spuriously increase N. I suppose you could
replace each missing data point by that person's mean ranking which
would be less problematic but it still wouldn't be a very good
approach.
I think you'd need ANOVA with multiple imputation or something like a
multilevel model unless the number of missing cases was small.
Thom
.
- References:
- Missing ranked data
- From: sxyvirgo
- Re: Missing ranked data
- From: Thom
- Missing ranked data
- Prev by Date: Re: Missing ranked data
- Next by Date: Uncertainty on Future Values of Y given measurements of X
- Previous by thread: Re: Missing ranked data
- Next by thread: Uncertainty on Future Values of Y given measurements of X
- Index(es):
Relevant Pages
|
